SkyPose: Real-time camera pose estimation via skyline matching in mountainous terrain

Abstract

Visual Positioning System (VPS) estimates a camera's pose from visual information of the surrounding environment and is widely used in autonomous driving, robotic navigation, and augmented reality (AR). Outdoor VPS commonly uses Global Navigation Satellite Systems (GNSS) and Inertial Measurement Units (IMU) as auxiliary cues. However, IMU accuracy can degrade due to magnetic disturbances and sensor noise. This paper presents a new outdoor VPS that precisely refines the IMU measurements by matching skylines extracted from input images with those generated from a Digital Elevation Model (DEM). The proposed system operates in real time and delivers high-accuracy pose estimates even under unreliable GNSS coverage and noisy IMU readings. The framework is robust and well-suited to mountainous terrain, enabling effective deployment in military, exploration, and AR-based applications.